Workshop 5: OpenAI Playground

Experiment with LLM parameters

What is the Playground?

The OpenAI Playground used to let you interact directly with GPT models, tweak parameters and see how these settings change the model's behavior.

It was discontinued in 2023, because OpenAI didn't want to expose the underlying model for safety reasons. For this workshop, I've built (vibe coded) a simplified version of the playground using ChatGPT instead of the original gpt-3 model.

The Tool

Go to The ChatGPT playground and enter the APÏ key

Key Parameters

Temperature (0-2): Controls randomness. 0 = deterministic, 2 = chaotic
Max tokens: Maximum length of the response
Top P: Alternative to temperature limits which tokens are considered
Model: Different ChatGPT models (gpt-4o, gpt-4o-mini, etc.)

Exercise 1: Temperature

Type whatever prompt and change only the temperature:

Set temperature to 0 — run it 3 times. What do you notice?
Set temperature to 1 — run it 3 times. What changes?
Set temperature to 2 — run it 3 times. What happens?

Exercise 2: Compare Models

Invent a hard prompt and try it with different models:

Try gpt-4o-mini — fast and cheap
Try gpt-4o — smarter but slower
Compare the quality, style, and response time

Try with: A logic puzzle, a coding question, or "Explain quantum entanglement like I'm 5"

Exercise 3: System Prompt

The system prompt sets the model's behavior. Try different ideas:

For example, use one of these system prompts:
- "You are a helpful assistant" (default)
- "You are a pirate. Respond only in pirate speak."
- "You are a skeptical scientist. Question everything."
- "Respond in exactly 10 words."

Questions to Consider

When would you want low temperature vs. high temperature?
Why might the same prompt give different results each time?
What's the trade-off between model size and cost/speed?
How does the system prompt change the "personality" without changing the model?
What happens if temperature is 0 but you run it twice? Why?